CN105718857A - Human body abnormal behavior detection method and system - Google Patents
Human body abnormal behavior detection method and system Download PDFInfo
- Publication number
- CN105718857A CN105718857A CN201610021212.7A CN201610021212A CN105718857A CN 105718857 A CN105718857 A CN 105718857A CN 201610021212 A CN201610021212 A CN 201610021212A CN 105718857 A CN105718857 A CN 105718857A
- Authority
- CN
- China
- Prior art keywords
- human body
- region
- action
- sequence
- deviant behavior
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
- Alarm Systems (AREA)
Abstract
The invention relates to a human body abnormal behavior detection method and system. The method comprises the following steps: according to video scene content, dividing a video scene into multiple areas, according to scene content included in each area, obtaining a motion set of motion which a human body can perform therein, and marking the performable motion sets in the multiple areas and each area; according to mark sets of the performable motion sets in the multiple areas and each area, obtaining abnormal behavior determination rules; obtaining a motion locus and a motion sequence of the human body in the multiple areas, and according to the mark sets, obtaining observable mark sequences of the motion locus and the motion sequence; and by use of the abnormal behavior determination rules, analyzing the observable mark sequences, determining whether the human body has abnormal behaviors, and if so, further determining the types of the abnormal behaviors. According to the invention, abnormal behaviors customized by supervisors and abnormal behaviors of the monitored people themselves can be comprehensively analyzed, and the identification rate is high.
Description
Technical field
The present invention relates to technical field of video monitoring, be specifically related to a kind of human body anomaly detection method and system.
Background technology
Video Supervision Technique is used in all trades and professions, under traditional monitoring mode, needs staff's moment to monitor the whether generation abnormal conditions of each camera scene on the one hand, it is easy to make sensory fatigue, thus causing missing inspection and flase drop.On the other hand, each camera continual shooting every day, store substantial amounts of video data, if the abnormal conditions of missing inspection require to look up corresponding video-frequency band, artificial lookup is relatively difficult, and inefficiency.Monitor video is carried out intellectual analysis and then becomes the research direction solving the problems referred to above.Existing intelligent monitor system carries out intellectual analysis mainly through detection and the moving target followed the tracks of in scene, also includes the behavior analysis to the moving target followed the tracks of.
At present, behavioral analysis technology is typically used in the fields such as Smart Home, financial industry, transportation industry and public safety.Such as in Smart Home industry, behavioral analysis technology can be used to the situations such as the falling down of monitor in real time old solitary people, stupor；In financial industry, can be used to occur between monitor in real time ATM fall, fight, the situation such as robbery；In transportation industry, can be used to the situations such as the traffic accident of some section of monitor in real time generation, personnel gather extremely；In public safety field, can be used to monitor the Mass disturbance such as the attack of terrorism, riot.
It is divided into two classes: a class is the self-defining Deviant Behavior of supervisor from the Deviant Behavior origin cause of formation; as line of stumbling, cross the border, drive in the wrong direction and stay; these Deviant Behavioies are not directly produced by monitored person; and interacting of monitored person and scene produces, supervisor carrys out these behaviors self-defined by monitored person's track in the scene or speed；Another kind of is the Deviant Behavior of monitored person self, as fought, falls down.The Deviant Behavior Producing reason of this two class is different, and most methods only have studied a part for this two classes Deviant Behavior, and a comprehensive method does not carry out overall these behaviors are studied.
Summary of the invention
For defect of the prior art, the present invention provides a kind of human body anomaly detection method and system, can to the self-defining Deviant Behavior of supervisor, Deviant Behavior and the Deviant Behavior of monitored person self that namely observed person and video scene interaction produce detect, improve the accuracy rate of Deviant Behavior identification, and then improve the reliability that Deviant Behavior is reported to the police.
First aspect, the invention provides a kind of human body anomaly detection method, including:
According to video scene content, this video scene is divided into multiple region, obtains human body enforceable set of actions wherein according to the scene content that each region comprises, and enforceable set of actions in the plurality of region and each region is carried out labelling；
Tag set according to implementing set of actions in the plurality of region and each region obtains Deviant Behavior judgment rule；
Obtain movement locus and the action sequence of human body in the plurality of region the observable labelled sequence according to the incompatible acquisition movement locus of described label sets and action sequence；
Utilize described Deviant Behavior judgment rule that described observable labelled sequence is analyzed, it is judged that whether above-mentioned human body has Deviant Behavior to occur, if so, determine whether to there occurs which kind of Deviant Behavior.
Alternatively, adopt numeral, character or the plurality of region of color mark, and adopt enforceable set of actions in numeral or each region of character marking.
Alternatively, following steps are adopted to generate the observable labelled sequence of movement locus, including:
Select a characteristic point as trace point from human body, obtain this trace point continuous path through one or more regions；
This continuous path is carried out sample code and generates discrete observable labelled sequence by the labelling according to above-mentioned continuous path region.
Alternatively, described characteristic point is the characteristic point in mass center of human body or foot.
Alternatively, described Deviant Behavior judgment rule includes Macroscopic Anomalies behavior judgment rule, and the spatial logic relation that described Macroscopic Anomalies behavior judgment rule is given between the distinctive function in the plurality of region and region by function or the supervisor in each region self obtains.
Alternatively, described Deviant Behavior judgment rule also includes microcosmic Deviant Behavior judgment rule；Described microcosmic Deviant Behavior judgment rule obtains according to the action sequence of monitored person self Deviant Behavior.
Alternatively, described method also includes the abnormal track sets in the human body Deviant Behavior video set adopting machine learning method training known and abnormal operation sequence, thus judging that whether unknown human body behavior is abnormal；Described machine learning method is one or more in support vector machines, iterative algorithm Adaboost and neutral net.
Alternatively, described method also includes adopting HMM HMM to be modeled human action identifying, including:
It is multiple attitude by each decomposition of movement of known human body；
Extract the characteristic vector of multiple attitudes of each action；
Adopt Kmeans clustering algorithm to be clustered by acquired each motion characteristic vector, obtain N number of cluster centre；
Calculate that each motion characteristic is vectorial and the Euclidean distance of each cluster centre, by each motion characteristic vector quantization to the cluster centre obtaining minimum Eustachian distance, to obtain the observable labelled sequence of the multiple attitude of each action；
Utilize the above-mentioned observable labelled sequence of Baum-Welch Algorithm for Training, thus obtaining the HMM model of each action；
Unknown human action being carried out feature extraction and quantization, is input in the HMM model of each corresponding action by obtained observable labelled sequence, this unknown action is under the jurisdiction of any action model to utilize forwards algorithms to judge.
Alternatively, also include adopting single characterization method or the assemblage characteristic method step to each motion characteristic vector optimization；Wherein,
Described single characterization method includes:
Respectively every one-dimensional characteristic is trained, as characteristic vector, the HMM model testing each action, thus obtaining each feature for the accuracy of the classification of motion and error rate, it is possible to feature the highest for error rate rejected；
Described assemblage characteristic method includes:
N is tieed up the dimension combination of the i in full feature and trains, as characteristic vector, the HMM model testing each action, wherein,Each i is all hadPlant number of combinations；
From test result, obtaining the accuracy of each combination, selecting maximum the combining as new full characteristic set of accuracy, thus feature little for contribution degree being rejected.
Second aspect, the embodiment of the present invention additionally provides a kind of human body unusual checking system, realizes based on human body anomaly detection method mentioned above, including:
Region and action mark unit, for this video scene being divided into multiple region according to video scene content, obtain human body enforceable set of actions wherein according to the scene content that each region comprises, and enforceable set of actions in the plurality of region and each region is carried out labelling；
Deviant Behavior judgment rule acquiring unit, for obtaining Deviant Behavior judgment rule according to the tag set that can implement set of actions in the plurality of region and each region；
Observable labelled sequence labelling acquiring unit, for obtaining movement locus and the action sequence of human body in the plurality of region, and the observable labelled sequence according to the incompatible acquisition movement locus of described label sets and action sequence；
Deviant Behavior judging unit, is used for utilizing described Deviant Behavior judgment rule that described observable labelled sequence is analyzed, it is judged that whether above-mentioned human body has Deviant Behavior to occur, and if so, determines whether to there occurs which kind of Deviant Behavior.
As shown from the above technical solution, the present invention can take into account the self-defining Deviant Behavior of supervisor, the Deviant Behavior of namely observed person and video scene interaction generation and the Deviant Behavior of monitored person self, improve the accuracy rate of Deviant Behavior identification, and then improve the reliability that Deviant Behavior is reported to the police.
Accompanying drawing explanation
Being more clearly understood from the features and advantages of the present invention by reference accompanying drawing, accompanying drawing is schematic and should not be construed as and the present invention is carried out any restriction, in the accompanying drawings:
Fig. 1 is a kind of human body anomaly detection method flow chart that the embodiment of the present invention provides；
Fig. 2～Fig. 4 is that the video scene region that the embodiment of the present invention provides divides and labelling schematic diagram；
Fig. 5 is the HMM model training process schematic of the known action that the embodiment of the present invention provides；
Fig. 6 provides image viewing feature schematic diagram in the training process that the embodiment of the present invention provides；
Fig. 7 is HMM model identification the unknown course of action schematic diagram that the embodiment of the present invention provides；
Fig. 8 is a kind of human body anomaly detection method flow chart in one embodiment of the invention；
Fig. 9 is a kind of human body unusual checking system block diagram that the embodiment of the present invention provides.
Detailed description of the invention
For making the purpose of the embodiment of the present invention, technical scheme and advantage clearly, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is a part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, the every other embodiment that those of ordinary skill in the art obtain under not making creative work premise, broadly fall into the scope of protection of the invention.
First aspect, the invention provides a kind of human body anomaly detection method, as it is shown in figure 1, include:
S100, according to video scene content, this video scene is divided into multiple region, obtain human body enforceable set of actions wherein according to the scene content that each region comprises, and enforceable set of actions in the plurality of region and each region is carried out labelling；
S200, according to the plurality of region and each region can be implemented set of actions tag set obtain Deviant Behavior judgment rule；
S300, the movement locus obtaining human body in the plurality of region and action sequence, and the observable labelled sequence according to the incompatible acquisition movement locus of described label sets and action sequence；
S400, utilize described Deviant Behavior judgment rule that described observable labelled sequence is analyzed, it is judged that whether above-mentioned human body has Deviant Behavior to occur, and if so, determines whether to there occurs which kind of Deviant Behavior.
Below each step of human body anomaly detection method provided by the invention is elaborated.First, introduce S100, according to video scene content, this video scene be divided into multiple region, obtain human body enforceable set of actions wherein according to the scene content that each region comprises, and enforceable set of actions in the plurality of region and each region is carried out the step of labelling.
The present invention is divided into some regions according to the content that video scene comprises.Can be realized by two kinds of methods it should be noted that scene partitioning is some regions: actual scenery that a kind of supervisor of being comprises according to scene or by the method for scenery color cluster in scene is manually divided.Such as road, desk, chair, door and window etc. all can divide or cluster and be one and close independent region, and the method has purposiveness.Another kind of method is that video scene is automatically divided into some grids by monitoring system, and each grid closes independent region as one, has blindness.For reducing time complexity, the embodiment of the present invention adopts numeral, character or different colors come each region of labelling, and adopt numeral or character to carry out each action of labelling.
Present example adopt supervisor manually video scene to be carried out zoning according to the actual scenery comprised in video scene.As shown in Figure 2 to 4, Fig. 2 is original scene, and Fig. 3 is zoning, and Fig. 4 is the scene mask obtained after dividing.Then analyze the set of actions that human body can be made in the scene, as walked, run, jump, squatting, crouch and wave.This example adopts capitalization (A-Z) to come labelling regional, each basic acts of lower case (a-z) labelling.
Secondly, introduce S200, obtain the step of Deviant Behavior judgment rule according to the tag set that can implement set of actions in the plurality of region and each region.
The present invention defines the behavior of human body from two levels of both macro and micro.Macro-level is that the track according to human body finite length in the scene judges that whether behavior is abnormal；Microcosmic level refers to that the action sequence with precedence according to human body finite length in the scene judges that whether its behavior is abnormal.More preferably, in the embodiment of the present invention, Deviant Behavior judgment rule includes Macroscopic Anomalies behavior judgment rule and microcosmic Deviant Behavior judgment rule.
Wherein, the spatial logic relation that molar behavior is the function according to each region self or supervisor gives between the distinctive function in the plurality of region and region is formulated.The function in each region self refers to that scene content can be supplied to the basic function (be used to sit, road be used to walk or race etc.) of people such as chair；Spatial logic relation between region refers to that human body can pass through two regions adjoined with certain speed under normal circumstances, it is impossible to trans-regional pass through；Supervisor gives the specific function in some region and refers to that supervisor is that in scene, some region defines the additional function (as road can only unidirectional be passed through, it is impossible to stay in certain region) except basic function.
It should be noted that Macroscopic Anomalies behavior in the present invention, the self-defining Deviant Behavior of supervisor all referring to monitored person from scene interactivity time due to the behavior defined with supervisor different and produce.Hereinafter, the present invention has made different titles at different paragraphs, " Macroscopic Anomalies behavior " in the present invention, " the self-defining Deviant Behavior of supervisor " and the Deviant Behavior of scene interactivity " monitored person with " belong to same concept, but just those skilled in the art will be appreciated that the application is the above-mentioned way in order to make those skilled in the art's understandings the technical program employing clearly.In like manner, hereinafter " microcosmic Deviant Behavior " and " Deviant Behavior of monitored person self " falls within is same concept.
For accurately judging Macroscopic Anomalies behavior, the embodiment of the present invention adopts following formula above-mentioned molar behavior is defined:
MacroActivity=(areaSymbolSet, startAreaSymbol, trackMinLength, trackMaxLength, macroGrammer)；
In formula, macroActivity refers to the result of microcosmic Deviant Behavior Macroscopic Anomalies behavior；AreaSymbolSet refers to the set in above-mentioned scene partitioning region；StartAreaSymbol refers to the human body initiation region when entering scene；TrackMinLength refers to minimum track sets length；TrackMaxLength refers to maximum track sets length.When the discrete loci sequence of sampling is more than or equal to trackMinLength, during less than or equal to trackMaxLength, system judges behavior character according to grammatical rules.When discrete loci sequence is more than trackMaxLength, adopting the data structure of queue, first in first out, discrete loci sequence length remains trackMaxLength.MacroGrammer refers to that microcosmic Deviant Behavior Macroscopic Anomalies behavior judges grammer.
In the embodiment of the present invention, Macroscopic Anomalies behavior generally includes following several:
1, (stop) is stayed: certain region residence time that target divides in scene or scene exceedes the time t that supervisor sets, then be judged to stay (stop)；
2, invasion (crossing the border): when target enters some off-limits region in scene, it is determined that for invasion；
3, drive in the wrong direction: when in the moving displacement direction of target and scene, the direction of motion angle of regulation is obtuse angle, it is determined that be retrograde；
4, hover: shuttle in some region that target divides in the scene, exceed the time T that supervisor sets, it is determined that for hovering；
5, trail: two to three targets movement locus in the scene is essentially identical, and these target trajectorys generate and have the regular hour poor, it is determined that for trailing.
6, extremely run: the movement velocity of target has exceeded the maximal rate of scene definition, it is determined that run for abnormal.
For accurately judging microcosmic Deviant Behavior, the embodiment of the present invention adopts following formula above-mentioned molecular behavior is defined:
MicroActivity=(actionSymbolSet, startActionSymbol, areaSymbol, activityMinLength, activityMaxLength, microGrammer)；
In formula, microActivity refers to the result of microcosmic level Deviant Behavior；ActionSymbolSet refers to the set of actions that human body can be made in the scene；StartActionSymbol refers to origination action during human body entrance scene；AreaSymbol refers to human body current kinetic region in the scene；ActivityMinLength refers to minimum action sequence length；ActivityMaxLength refers to maximum action sequence length；MicroGrammer refers to that microcosmic level Deviant Behavior judges grammer.
In the present invention, molecular behavior rule is formulated according to the action sequence of monitored person self Deviant Behavior.Microcosmic Deviant Behavior includes following several:
1, fall: target action sequence in the scene from walking or run to become crouching, is then now in alert status, exceed, when target is sleeping, the time t that supervisor sets in the scene, be then judged to collapse in the ground.
2 fight: multiple targets in the scene, action sequence acute variation, and punch occur, kick, the action such as wave, it is determined that for fighting.
In practical application, microGrammer can adopt two kinds of formulating methods: 1) supervisor directly formulates according to the action sequence of specific exceptions behavior to be detected in scene；2) according to the video set of continuous action in known scene, obtained the grader of various actions by machine learning training, then bring the action sequence of unknown video into grader and carry out behavior judgement.
Again, introduce S300, obtain movement locus and the action sequence of human body in the plurality of region, and the step according to the incompatible acquisition movement locus of described label sets and the observable labelled sequence of action sequence.
In the embodiment of the present invention, when human body has divided the scene areas of also labelling at ceaselessly movement thereacross, in real time detection region residing for human body, using the labelling in this region observable labelling as human body motion track, records in track labelled sequence in order by mark value；When human body rests in certain region, define dwell period T1, when time of staying t=nT1 (n=1,2,3 ...), can be regarded as once " passing through ", by the mark value record in this region in track labelled sequence.
It should be noted that, in the embodiment of the present invention, first have to utilize the algorithm of machine learning that human action is trained, and obtain the grader of each action.Human body is after entering scene, definition action collection period T2, the image set in each cycle is input in described classification of motion device according to sequencing, thus identifying the action of current human, and by labelling record corresponding in action mark set for this action in action mark sequence.
Alternatively, the present invention adopts Hidden Markov (HMM) model the study of each parsing action modeling is identified, for instance a HMM model is made up of 5 parameters: λ=(S, O, T, E, π)；Wherein, S is hidden state set, and O is observable assemble of symbol, and both set numbers can define according to application scenarios, and T is hidden state transition probability matrix, and E is symbol emission probability matrix, and π is initial hidden probability distribution over states.
As it is shown in figure 5, during training, bring in above-mentioned HMM model by the observable labelled sequence of same behavior, according to Baum-Welch algorithm, the value of training two matrixes of T, E, model λ can be obtained by maximizing probability P (O │ λ).The corresponding HMM model of each behavior in scene, utilizes the training data study of each behavior to obtain model set { λ1, λ2..., λi... }, here it is training process.For the observable labelled sequence of a unknown molecular behavior, calculate its likelihood score P with each HMM (O | λi).By this unknown molecular behavior likelihood score P (O | λi) in maximum behavior classification, action recognition can be realized.
In the embodiment of the present invention, the HMM model training process of known action is as it is shown in figure 5, detailed process includes: read the video sequence of current training action, current frame image is carried out gray processing process；The modeling of background and renewal, prospect is asked for by background subtraction, by the opening and closing operation denoising to image, thus finding the profile of independent connected region and this region, profile is carried out target recognition and tracking, thus finding the profile belonging to human body, these profiles are carried out multidimensional characteristic vectors extraction, the all characteristic vectors obtained are clustered, the quantity of cluster is determined by the element number of the observable labelled sequence collection of HMM, then each characteristic vector is planned in affiliated classification, thus obtaining observable sequence, bring in HMM model, training E, T matrix, obtain maximum likelihood degree.
In practical application, the picture format of video is RGB or YUV.Therefore, a certain for video two field picture gray processing is processed, gray level image is converted to for RGB or yuv space image, it is possible to adopt conventional following two formula:
Gray1=R*0.299+G*0.587+B*0.114；
Gray2=Y；
Certainly, those skilled in the art can also adopt integer arithmetic or integer shift operation according to practical situation, and the present invention is not construed as limiting.
In the embodiment of the present invention, become static situation in order to effectively reduce noise and detection human body from motion, adopt double; two background subtraction to detect foreground moving and static target and the method for double; two context update.
The embodiment of the present invention defines two kinds of background models.One is fast background, and one is slow background.Consider that in a lot of situation, change of background has seriality, in order to simplify calculating, method proposes a kind of fast background Bf(x, update mechanism y):
Wherein, (x y) represents the coordinate position of video image；Represent the fast background of nth frame；I1(x y) represents the image after video the 1st frame gray processing；Represent the fast background of N-1 frame；IN(x y) represents the image after video nth frame gray processing；α is learning rate, and this value is more big, and context update is more fast.Often reading a frame video, fast background all can update once.
This method also proposed a kind of slow background Bs (x, update mechanism y):
Wherein,Represent the slow background of n-th frame；Represent the slow background of n-1 frame；N represents the nth frame of video；K represents turnover rate, and namely every k frame updates once slow background, and its value can adjust according to the quality actually obtaining slow background effect；Represent the fast background of nth frame；TdRepresent threshold value, when the absolute difference of fast background and slow background respective coordinates point is less than this value, directly update slow background with the pixel value of fast background.Otherwise, with the slow background of previous frame respective coordinates point pixel value update present frame slow background.
The present invention detects foreground moving object: adopts the update method of fast background by learning in fast background model by moving to static target, the gray-scale map of video present frame and fast background is done difference, and arranges threshold filtering, can obtain the foreground mask M of moving targetm.Equally, the update method of slow background has only learnt static scene, the gray-scale map of video present frame and slow background do difference, and arrange threshold filtering, can obtain the foreground mask M of motion and static targeta.Utilize the foreground mask M of motion and static targetaDeduct the foreground mask M of moving targetmThe foreground mask M of static target can be obtaineds, by these masks and video original image are done the static and moving target that intersection operation can obtain in prospect.
For realizing removing due to the connected region of the formation of noise, leaving the target of the connected region of moving object and profile in scene, the present invention uses morphologic filtering that image carries out opening operation and closed operation.It should be noted that the embodiment of the present invention uses prior art to carry out target recognition and tracking, wherein adopt histograms of oriented gradients (HistogramofOrientedGradient, HOG) to detect pedestrian, adopt meanshift algorithm to carry out target following.If the moving target in scene is defined as human body, then can save pedestrian detection process and directly follow the tracks of target.
For finding the key feature that contribution degree that behavior is classified is big, the present embodiment extracts 30 dimensional feature vectors of image.Fig. 6 provides image viewing feature schematic diagram in the training process that the embodiment of the present invention provides.The impact far and near in order to reduce video camera, as shown in Figure 6, the present invention utilizes minimum enclosed rectangle (or oval) by characteristic standardization.Minimum enclosed rectangle (or oval) is the rectangle (or oval) of the minimum area surrounding foreground target.According to extraction step, these features can be divided into four classifications:
1, framework characteristic extracts, it is possible to be called that the contour feature of N sector region extracts.At the barycenter place of foreground target, target 360 ° is divided into N number of sector region.In each sector region, calculate barycenter to the ultimate range of object boundary, afterwards by this distance catercorner length (or long axis length of ellipse) divided by minimum enclosed rectangle, to be standardized.In the present embodiment, foreground target is divided into 8 sector regions, thus extracts 1～8 dimensional feature.
2, area distributions feature extraction.By the N number of sector region obtained in previous step, add up the area of foreground target in each sector region, namely the number of the pixel of foreground target, then the area divided by whole foreground target carrys out standardization.In the present embodiment, foreground target is divided into 8 sector regions, thus obtaining 9～16 dimensional features.
3, Motion feature extraction.In practical application, main shaft can approximate representation motion time the information such as the angle of human body, center of gravity and symmetry, and to insensitive for noise, therefore using this main shaft angle as the 17th dimensional feature.Main shaft angle computing formula is as follows:
In formula, M11、M20And M02It is the second moment in foreground target region.
4, the speed of barycenter:
Barycenter acceleration:
Wherein, vx、ax、s1x、s2xRepresent respectively and be parallel to the speed of image x-axis, acceleration and displacement component；Vy、ay、s1y、s2yRepresent respectively and be parallel to the speed of image y-axis, acceleration and displacement component, and constitute 18～21 dimensional features.
5, geometry appearance features.Depth-width ratio refers to, the depth-width ratio (or ratio of semi-minor axis length of ellipse) of the minimum area-encasing rectangle of foreground target；Filling rate refers to, the area ratio of foreground target and minimum area-encasing rectangle (or oval).In the embodiment of the present invention, depth-width ratio and filling rate constitute 22～23 dimensional features.6, rotation invariant moment.Calculate the image 7 the Hu square to position, convergent-divergent and invariable rotary, constitute 24～30 dimensional features.
It should be noted that, the embodiment of the present invention only illustrates Partial Feature and the extraction process thereof of representative image key feature, it will be obvious to one with ordinary skill in the art that such as textural characteristics also can as the key feature of image, certain those skilled in the art can also select suitable key feature to realize the classification to image according to specifically used scene, and the present invention is not construed as limiting.
For the observable labelled sequence of 30 dimensional feature vectors of above-mentioned image Yu HMM model is united, it is necessary to carry out clustering to this 30 dimensional feature vector and quantify.Such as, the embodiment of the present invention adopts Kmeans algorithm that this 30 dimensional feature vector is clustered, and obtains N number of cluster centre, and N is cluster categorical measure, and its value depends on the span of the observable labelled sequence of HMM model.Then pass through the Euclidean distance calculating each characteristic vector with each cluster centre, thus each characteristic vector can be quantified to the cluster centre obtaining minimum Eustachian distance, to obtain one group of observable labelled sequence.
The embodiment of the present invention adopt Baum-Welch algorithm train the parameter of HMM model.Using all observable labelled sequence of the same behavior input quantity as Baum-Welch algorithm, T, E matrix in training HMM model, by maximizing probability P (Oi|λi), thus obtaining model λi.It is respectively trained and often organizes behavioral data, finally give the model set { λ of all HMM model1, λ2..., λi,…}。
As it is shown in fig. 7, the process of the HMM model identification the unknown action after adopting training in the embodiment of the present invention includes: gather video sequence, acquired image is carried out gray processing.Then to background modeling and renewal, background subtraction prospect is asked for, by the opening and closing operation denoising to image, thus finding the profile of independent connected region and this region, profile being carried out target recognition and tracking, thus finding the profile belonging to human body, obtaining the multidimensional characteristic vectors of human body.It should be noted that in action recognition process, it is not necessary to again to feature vector clusters, only the cluster centre in training process need to be utilized to quantify.Then the characteristic sequence that unknown action video proposes is substituted into the model set { λ of HMM model1, λ2..., λi... } in, use forwards algorithms to try to achieve maximum P (Oi|λi), this video sequence is then identified as model λiCorresponding behavior.
It will be appreciated that the contribution degree of the classification of motion is different by each dimensional feature vector in 30 dimensional feature vectors extracted in the embodiment of the present invention.Therefore, features described above set is also optimized by the present invention, the characteristic vector that during to reject the classification of motion, contribution degree is little, so can improve the real-time that characteristic vector calculates.
For achieving the above object, the embodiment of the present invention adopts single characteristic test method and assemblage characteristic method of testing totally two kinds of methods.Wherein, single characteristic test method refers to, is respectively adopted every one-dimensional characteristic and makes to train the HMM model of each action, such that it is able to obtain each dimensional feature vector for the accuracy of the classification of motion and error rate.Then characteristic vector the highest for error rate is rejected.Assemblage characteristic method of testing refers to, N ties up the dimension assemblage characteristic of the i in full feature and trains, as characteristic vector, the HMM model testing each action, wherein,Each i assemblage characteristic is all hadPlant number of combinations；From test result, obtaining the accuracy of each combination, selecting maximum the combining as new full characteristic set of accuracy, thus feature little for contribution degree being rejected.
Finally, S400, utilize described Deviant Behavior judgment rule that described observable labelled sequence is analyzed, it is judged that whether above-mentioned human body has Deviant Behavior to occur, and if so, determines whether to there occurs which kind of Deviant Behavior.
Many groups observable labelled sequence acquired in the embodiment of the present invention includes molar behavior observable labelled sequence and molecular behavior observable labelled sequence.The kind judging Deviant Behavior the process reported to the police is set forth separately below with Macroscopic Anomalies behavior and microcosmic Deviant Behavior example.
As shown in Figure 8, the present invention is left example with Macroscopic Anomalies behavioral value with funny: the first scene partitioning region areaSymbolSet in definition scene, it is assumed that include A, B, five regions of C, D and E, and wherein supervisor arranges C region is non-stay district.Assuming that monitored person enters the prime area startAreaSymbol of scene is B region, minimum track sets length trackMinLength is 5, maximum track sets length trackMaxLength is 10, that stays judges that grammer macroGrammer is described as: when the track sets of monitored person occur continuously 5 times and above tab character as B time, track labelled sequence such as monitored person is: CCCCBBBBBB, then be judged to stay.
Microcosmic unusual checking is to fall: the first set of actions actionSymbolSet in definition scene, it is assumed that including: walk (being labeled as a), run (being labeled as b), jump (being labeled as c), squat (being labeled as d), sit (being labeled as e) and sleeping (being labeled as f) etc.；Defining minimum recognizable sequence length activityMinLength is 5, namely only collects monitored person and enters in scene and just start to identify its behavior after 5 actions；Defining maximum discernible action sequence length activityMaxLength is 10, if the action sequence of monitored person is not up to this value, then continue to record its action, if it exceeds this value, then the action the earliest of time in action sequence being removed, the afterbody that new element is added to this action sequence adds.
Assume that monitored person enters the origination action startActionSymbol of scene areas and walks；Its residing region areaSymbol is walking district, and it is labeled as A；And specify that actions such as sitting and crouch can not be made in this region.The cycle of action collection is T, and per elapsed time T gathers monitored person's action and adds action sequence.Fall judge grammer microGrammer be described as when the action sequence of monitored person be all sleeping or seat time, it is determined that for falling.According to above-mentioned judgement grammer monitored person in walking district by sleeping can be described as of walking:
{ A │ a, a, a, a, a, a, f, f, f, f}；
Namely monitored person is in walking district (A), and front 6 actions are away (a), and rear four actions are sleeping (f)；Now early warning monitored person falls down, and has two kinds of situations afterwards: one is that monitored person stands up voluntarily, then need not report to the police；Another kind be monitored person unconscious or station not, then produce the warning message fallen.When monitored person continues to gather its action sequence after falling down, at sleeping (f) or sit and produce when whole action sequence is filled up in the action of (e) to report to the police (such as { A │ ffffffeeee}), if unfilled action sequence, and action sequence occurs (a) (such as { A │ ffffeeeeaa}), then do not produce to report to the police and eliminate early warning.
For embodying the superiority of a kind of human body anomaly detection method that the embodiment of the present invention provides, the embodiment of the present invention additionally provides a kind of human body unusual checking system, as it is shown in figure 9, include:
Region and action mark unit, for this video scene being divided into multiple region according to video scene content, obtain human body enforceable set of actions wherein according to the scene content that each region comprises, and enforceable set of actions in the plurality of region and each region is carried out labelling；
Deviant Behavior judgment rule acquiring unit, for obtaining Deviant Behavior judgment rule according to the tag set that can implement set of actions in the plurality of region and each region；
Observable labelled sequence acquiring unit, for obtaining movement locus and the action sequence of human body in the plurality of region, and the observable labelled sequence according to the incompatible acquisition movement locus of label sets and action sequence that can implement set of actions in the plurality of region and each region；
Deviant Behavior judging unit, is used for utilizing described Deviant Behavior judgment rule that described observable labelled sequence is analyzed, it is judged that whether above-mentioned human body has Deviant Behavior to occur, and if so, determines whether to there occurs which kind of Deviant Behavior.
Human body unusual checking system provided by the invention realizes based on human body anomaly detection method mentioned above, thus can solve same technical problem, and obtains identical technique effect, and this is no longer going to repeat them.
In sum, human body anomaly detection method provided by the invention and system, by video scene being carried out partition and enforceable action in multiple regions and each region being carried out labelling, obtain Deviant Behavior judgment rule according to zone marker and action mark set；Then human body movement locus in multiple regions and action observable labelled sequence are obtained；Utilize above-mentioned Deviant Behavior judgment rule that observable labelled sequence is analyzed.The present invention can take into account the self-defining Deviant Behavior of supervisor, the Deviant Behavior of namely observed person and video scene interaction generation and the Deviant Behavior of monitored person self, improve the accuracy rate of Deviant Behavior identification, and then improve the reliability that Deviant Behavior is reported to the police.
In the present invention, term " first ", " second ", " the 3rd " only for descriptive purposes, and it is not intended that instruction or hint relative importance.Term " multiple " refers to two or more, unless otherwise clear and definite restriction.Although being described in conjunction with the accompanying embodiments of the present invention, but those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such amendment and modification each fall within the scope being defined by the appended claims.
Claims (10)
1. a human body anomaly detection method, it is characterised in that including:
According to video scene content, this video scene is divided into multiple region, obtains human body enforceable set of actions wherein according to the scene content that each region comprises, and enforceable set of actions in the plurality of region and each region is carried out labelling；
Tag set according to implementing set of actions in the plurality of region and each region obtains Deviant Behavior judgment rule；
Obtain movement locus and the action sequence of human body in the plurality of region the observable labelled sequence according to the incompatible acquisition movement locus of described label sets and action sequence；
Utilize described Deviant Behavior judgment rule that described observable labelled sequence is analyzed, it is judged that whether above-mentioned human body has Deviant Behavior to occur, if so, determine whether to there occurs which kind of Deviant Behavior.
2. human body anomaly detection method according to claim 1, it is characterised in that adopt numeral, character or the plurality of region of color mark, and adopt enforceable set of actions in numeral or each region of character marking.
3. human body anomaly detection method according to claim 2, it is characterised in that adopt following steps to generate the observable labelled sequence of movement locus, including:
Select a characteristic point as trace point from human body, obtain this trace point continuous path through one or more regions；
This continuous path is carried out sample code and generates discrete observable labelled sequence by the labelling according to above-mentioned continuous path region.
4. human body anomaly detection method according to claim 3, it is characterised in that described characteristic point is the characteristic point in mass center of human body or foot.
5. the human body anomaly detection method according to Claims 1 to 4 any one, it is characterized in that, described Deviant Behavior judgment rule includes Macroscopic Anomalies behavior judgment rule, and the spatial logic relation that described Macroscopic Anomalies behavior judgment rule is given between the distinctive function in the plurality of region and region by function or the supervisor in each region self obtains.
6. the human body anomaly detection method according to Claims 1 to 4 any one, it is characterised in that described Deviant Behavior judgment rule also includes microcosmic Deviant Behavior judgment rule；Described microcosmic Deviant Behavior judgment rule obtains according to the action sequence of monitored person self Deviant Behavior.
7. the human body anomaly detection method according to Claims 1 to 4 any one, it is characterized in that, described method also includes the abnormal track sets in the human body Deviant Behavior video set adopting machine learning method training known and abnormal operation sequence, thus judging that whether unknown human body behavior is abnormal；Described machine learning method is one or more in support vector machines, iterative algorithm Adaboost and neutral net.
8. human body anomaly detection method according to claim 7, it is characterised in that described method also includes adopting HMM HMM to be modeled human action identifying, including:
It is multiple attitude by each decomposition of movement of known human body；
Extract the characteristic vector of multiple attitudes of each action；
Adopt Kmeans clustering algorithm to be clustered by acquired each motion characteristic vector, obtain N number of cluster centre；
Calculate that each motion characteristic is vectorial and the Euclidean distance of each cluster centre, by each motion characteristic vector quantization to the cluster centre obtaining minimum Eustachian distance, to obtain the observable labelled sequence of the multiple attitude of each action；
Utilize the above-mentioned observable labelled sequence of Baum-Welch Algorithm for Training, thus obtaining the HMM model of each action；
Unknown human action being carried out feature extraction and quantization, is input in the HMM model of each corresponding action by obtained observable labelled sequence, this unknown action is under the jurisdiction of any action model to utilize forwards algorithms to judge.
9. human body anomaly detection method according to claim 8, it is characterised in that also include adopting single characterization method or the assemblage characteristic method step to each motion characteristic vector optimization；Wherein,
Described single characterization method includes:
Respectively every one-dimensional characteristic is trained, as characteristic vector, the HMM model testing each action, thus obtaining each feature for the accuracy of the classification of motion and error rate, it is possible to feature the highest for error rate rejected；
Described assemblage characteristic method includes:
N is tieed up the dimension combination of the i in full feature and trains, as characteristic vector, the HMM model testing each action, wherein,Each i is all hadPlant number of combinations；
From test result, obtaining the accuracy of each combination, selecting maximum the combining as new full characteristic set of accuracy, thus feature little for contribution degree being rejected.
10. a human body unusual checking system, realizes based on the human body anomaly detection method described in claim 1～9 any one, it is characterised in that including:
Region and action mark unit, for this video scene being divided into multiple region according to video scene content, obtain human body enforceable set of actions wherein according to the scene content that each region comprises, and enforceable set of actions in the plurality of region and each region is carried out labelling；
Deviant Behavior judgment rule acquiring unit, for obtaining Deviant Behavior judgment rule according to the tag set that can implement set of actions in the plurality of region and each region；
Observable labelled sequence labelling acquiring unit, for obtaining movement locus and the action sequence of human body in the plurality of region, and the observable labelled sequence according to the incompatible acquisition movement locus of described label sets and action sequence；
Deviant Behavior judging unit, is used for utilizing described Deviant Behavior judgment rule that described observable labelled sequence is analyzed, it is judged that whether above-mentioned human body has Deviant Behavior to occur, and if so, determines whether to there occurs which kind of Deviant Behavior.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610021212.7A CN105718857B (en) | 2016-01-13 | 2016-01-13 | A kind of human body anomaly detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610021212.7A CN105718857B (en) | 2016-01-13 | 2016-01-13 | A kind of human body anomaly detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105718857A true CN105718857A (en) | 2016-06-29 |
| CN105718857B CN105718857B (en) | 2019-06-04 |
Family
ID=56147821
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610021212.7A Active CN105718857B (en) | 2016-01-13 | 2016-01-13 | A kind of human body anomaly detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105718857B (en) |
Cited By (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106128022A (en) * | 2016-07-18 | 2016-11-16 | 四川君逸数码科技股份有限公司 | A kind of wisdom gold eyeball identification violent action alarm method and device |
| CN106128053A (en) * | 2016-07-18 | 2016-11-16 | 四川君逸数码科技股份有限公司 | A kind of wisdom gold eyeball identification personnel stay hover alarm method and device |
| CN106228572A (en) * | 2016-07-18 | 2016-12-14 | 西安交通大学 | The long inactivity object detection of a kind of carrier state mark and tracking |
| CN106373336A (en) * | 2016-08-30 | 2017-02-01 | 苏州品诺维新医疗科技有限公司 | Fall detection method and device |
| CN107036596A (en) * | 2017-04-12 | 2017-08-11 | 无锡研测技术有限公司 | Industrial bracelet based on MEMS inertial sensor module |
| CN107480618A (en) * | 2017-08-02 | 2017-12-15 | 深圳微品时代网络技术有限公司 | A kind of data analysing method of big data platform |
| CN107688828A (en) * | 2017-08-23 | 2018-02-13 | 武汉大学 | A kind of bus degree of crowding estimating and measuring method based on mobile phone sensor |
| CN107914124A (en) * | 2016-10-07 | 2018-04-17 | 发那科株式会社 | Operation auxiliary system with rote learning portion |
| CN108062521A (en) * | 2017-12-12 | 2018-05-22 | 深圳大学 | Method for detecting human face, device, terminal and medium based on convolutional neural networks |
| CN108229407A (en) * | 2018-01-11 | 2018-06-29 | 武汉米人科技有限公司 | A kind of behavioral value method and system in video analysis |
| CN108256540A (en) * | 2016-12-28 | 2018-07-06 | ***通信有限公司研究院 | A kind of information processing method and system |
| CN109064686A (en) * | 2018-08-17 | 2018-12-21 | 浙江捷尚视觉科技股份有限公司 | A kind of ATM trailing detection method based on human body segmentation |
| CN109086731A (en) * | 2018-08-15 | 2018-12-25 | 深圳市烽焌信息科技有限公司 | It is a kind of for carrying out the robot and storage medium of behavior monitoring |
| CN109145804A (en) * | 2018-08-15 | 2019-01-04 | 深圳市烽焌信息科技有限公司 | Behavior monitoring method and robot |
| CN109214292A (en) * | 2018-08-06 | 2019-01-15 | 广东技术师范学院 | A kind of picked angle recognition method and apparatus of human body based on BP neural network |
| CN109298646A (en) * | 2018-11-09 | 2019-02-01 | 珠海格力电器股份有限公司 | Scene control method and device, storage medium and electric appliance |
| CN109299646A (en) * | 2018-07-24 | 2019-02-01 | 北京旷视科技有限公司 | Crowd's accident detection method, apparatus, system and storage medium |
| CN109325474A (en) * | 2018-11-14 | 2019-02-12 | 郭道宁 | A kind of abnormal state detection method of couple of special caregiver of need |
| CN109344705A (en) * | 2018-08-27 | 2019-02-15 | 广州烽火众智数字技术有限公司 | A kind of pedestrian behavior detection method and system |
| CN109558865A (en) * | 2019-01-22 | 2019-04-02 | 郭道宁 | A kind of abnormal state detection method to the special caregiver of need based on human body key point |
| CN109584490A (en) * | 2018-12-10 | 2019-04-05 | Tcl通力电子（惠州）有限公司 | Safety protection method, intelligent sound box and security system |
| CN109684916A (en) * | 2018-11-13 | 2019-04-26 | 恒睿(重庆)人工智能技术研究院有限公司 | Based on path locus data exception detection method, system, equipment and storage medium |
| CN109815921A (en) * | 2019-01-29 | 2019-05-28 | 北京融链科技有限公司 | The prediction technique and device of the class of activity in hydrogenation stations |
| CN109993037A (en) * | 2018-01-02 | 2019-07-09 | ***通信有限公司研究院 | Action identification method, device, wearable device and computer readable storage medium |
| CN110062210A (en) * | 2019-04-28 | 2019-07-26 | 广东安居宝数码科技股份有限公司 | Monitoring method, device, equipment and the storage medium of object |
| CN110399437A (en) * | 2018-04-23 | 2019-11-01 | 北京京东尚科信息技术有限公司 | Behavior analysis method and device, electronic equipment, storage medium |
| CN110414360A (en) * | 2019-07-02 | 2019-11-05 | 桂林电子科技大学 | A kind of detection method and detection device of abnormal behaviour |
| CN110557676A (en) * | 2018-05-30 | 2019-12-10 | 罗技欧洲公司 | System and method for determining and recommending video content active areas for a scene |
| CN110555394A (en) * | 2019-08-19 | 2019-12-10 | 西安理工大学 | Fall risk assessment method based on human body shape characteristics |
| WO2020001216A1 (en) * | 2018-06-26 | 2020-01-02 | 杭州海康威视数字技术股份有限公司 | Abnormal event detection |
| CN110705505A (en) * | 2019-10-14 | 2020-01-17 | 国网山东省电力公司泰安供电公司 | Intelligent identification device, method and system for service specification of electric power business hall |
| CN111062356A (en) * | 2019-12-26 | 2020-04-24 | 沈阳理工大学 | Method for automatically identifying human body action abnormity from monitoring video |
| CN111652201A (en) * | 2020-08-10 | 2020-09-11 | 中国人民解放军国防科技大学 | Video data abnormity identification method and device based on depth video event completion |
| CN112017403A (en) * | 2020-09-15 | 2020-12-01 | 青岛联合创智科技有限公司 | Community-house integrated intelligent service electronic board |
| CN112183153A (en) * | 2019-07-01 | 2021-01-05 | ***通信集团浙江有限公司 | Object behavior detection method and device based on video analysis |
| CN112382032A (en) * | 2020-11-13 | 2021-02-19 | 珠海大横琴科技发展有限公司 | Monitoring method and device, electronic equipment and storage medium |
| CN113177459A (en) * | 2021-04-25 | 2021-07-27 | 云赛智联股份有限公司 | Intelligent video analysis method and system for intelligent airport service |
| CN113505713A (en) * | 2021-07-16 | 2021-10-15 | 上海塞嘉电子科技有限公司 | Intelligent video analysis method and system based on airport security management platform |
| CN113705355A (en) * | 2021-07-30 | 2021-11-26 | 汕头大学 | Real-time detection method for abnormal behaviors |
| CN113705274A (en) * | 2020-05-20 | 2021-11-26 | 杭州海康威视数字技术股份有限公司 | Climbing behavior detection method and device, electronic equipment and storage medium |
| CN114419528A (en) * | 2022-04-01 | 2022-04-29 | 浙江口碑网络技术有限公司 | Anomaly identification method and device, computer equipment and computer readable storage medium |
| CN116434346A (en) * | 2023-06-12 | 2023-07-14 | 四川汉唐云分布式存储技术有限公司 | Method and device for detecting customer behaviors in unattended store and storage medium |
| CN116430831A (en) * | 2023-04-26 | 2023-07-14 | 宁夏五谷丰生物科技发展有限公司 | Data abnormity monitoring method and system applied to edible oil production control system |
| KR102591483B1 (en) * | 2022-11-16 | 2023-10-19 | 주식회사우경정보기술 | Apparatus and method for spatiotemporal neural network-based labeling for building anomaly data set |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101271527A (en) * | 2008-02-25 | 2008-09-24 | 北京理工大学 | Exception action detecting method based on athletic ground partial statistics characteristic analysis |
| CN101719216A (en) * | 2009-12-21 | 2010-06-02 | 西安电子科技大学 | Movement human abnormal behavior identification method based on template matching |
| CN101872418A (en) * | 2010-05-28 | 2010-10-27 | 电子科技大学 | Detection method based on group environment abnormal behavior |
| US20120314064A1 (en) * | 2011-06-13 | 2012-12-13 | Sony Corporation | Abnormal behavior detecting apparatus and method thereof, and video monitoring system |
| CN103390278A (en) * | 2013-07-23 | 2013-11-13 | 中国科学技术大学 | Detecting system for video aberrant behavior |
-
2016
- 2016-01-13 CN CN201610021212.7A patent/CN105718857B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101271527A (en) * | 2008-02-25 | 2008-09-24 | 北京理工大学 | Exception action detecting method based on athletic ground partial statistics characteristic analysis |
| CN101719216A (en) * | 2009-12-21 | 2010-06-02 | 西安电子科技大学 | Movement human abnormal behavior identification method based on template matching |
| CN101872418A (en) * | 2010-05-28 | 2010-10-27 | 电子科技大学 | Detection method based on group environment abnormal behavior |
| US20120314064A1 (en) * | 2011-06-13 | 2012-12-13 | Sony Corporation | Abnormal behavior detecting apparatus and method thereof, and video monitoring system |
| CN103390278A (en) * | 2013-07-23 | 2013-11-13 | 中国科学技术大学 | Detecting system for video aberrant behavior |
Cited By (53)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106128053A (en) * | 2016-07-18 | 2016-11-16 | 四川君逸数码科技股份有限公司 | A kind of wisdom gold eyeball identification personnel stay hover alarm method and device |
| CN106228572A (en) * | 2016-07-18 | 2016-12-14 | 西安交通大学 | The long inactivity object detection of a kind of carrier state mark and tracking |
| CN106128022A (en) * | 2016-07-18 | 2016-11-16 | 四川君逸数码科技股份有限公司 | A kind of wisdom gold eyeball identification violent action alarm method and device |
| CN106228572B (en) * | 2016-07-18 | 2019-01-29 | 西安交通大学 | A kind of the long inactivity object detection and tracking of carrier state mark |
| CN106373336A (en) * | 2016-08-30 | 2017-02-01 | 苏州品诺维新医疗科技有限公司 | Fall detection method and device |
| CN107914124A (en) * | 2016-10-07 | 2018-04-17 | 发那科株式会社 | Operation auxiliary system with rote learning portion |
| CN108256540A (en) * | 2016-12-28 | 2018-07-06 | ***通信有限公司研究院 | A kind of information processing method and system |
| CN107036596A (en) * | 2017-04-12 | 2017-08-11 | 无锡研测技术有限公司 | Industrial bracelet based on MEMS inertial sensor module |
| CN107480618A (en) * | 2017-08-02 | 2017-12-15 | 深圳微品时代网络技术有限公司 | A kind of data analysing method of big data platform |
| CN107688828B (en) * | 2017-08-23 | 2020-11-03 | 武汉大学 | Bus crowding degree estimation method based on mobile phone sensor |
| CN107688828A (en) * | 2017-08-23 | 2018-02-13 | 武汉大学 | A kind of bus degree of crowding estimating and measuring method based on mobile phone sensor |
| CN108062521A (en) * | 2017-12-12 | 2018-05-22 | 深圳大学 | Method for detecting human face, device, terminal and medium based on convolutional neural networks |
| CN109993037A (en) * | 2018-01-02 | 2019-07-09 | ***通信有限公司研究院 | Action identification method, device, wearable device and computer readable storage medium |
| CN108229407A (en) * | 2018-01-11 | 2018-06-29 | 武汉米人科技有限公司 | A kind of behavioral value method and system in video analysis |
| CN110399437A (en) * | 2018-04-23 | 2019-11-01 | 北京京东尚科信息技术有限公司 | Behavior analysis method and device, electronic equipment, storage medium |
| CN110557676A (en) * | 2018-05-30 | 2019-12-10 | 罗技欧洲公司 | System and method for determining and recommending video content active areas for a scene |
| WO2020001216A1 (en) * | 2018-06-26 | 2020-01-02 | 杭州海康威视数字技术股份有限公司 | Abnormal event detection |
| CN109299646B (en) * | 2018-07-24 | 2021-06-25 | 北京旷视科技有限公司 | Crowd abnormal event detection method, device, system and storage medium |
| CN109299646A (en) * | 2018-07-24 | 2019-02-01 | 北京旷视科技有限公司 | Crowd's accident detection method, apparatus, system and storage medium |
| CN109214292A (en) * | 2018-08-06 | 2019-01-15 | 广东技术师范学院 | A kind of picked angle recognition method and apparatus of human body based on BP neural network |
| CN109145804A (en) * | 2018-08-15 | 2019-01-04 | 深圳市烽焌信息科技有限公司 | Behavior monitoring method and robot |
| CN109086731A (en) * | 2018-08-15 | 2018-12-25 | 深圳市烽焌信息科技有限公司 | It is a kind of for carrying out the robot and storage medium of behavior monitoring |
| CN109064686A (en) * | 2018-08-17 | 2018-12-21 | 浙江捷尚视觉科技股份有限公司 | A kind of ATM trailing detection method based on human body segmentation |
| CN109344705A (en) * | 2018-08-27 | 2019-02-15 | 广州烽火众智数字技术有限公司 | A kind of pedestrian behavior detection method and system |
| CN109298646A (en) * | 2018-11-09 | 2019-02-01 | 珠海格力电器股份有限公司 | Scene control method and device, storage medium and electric appliance |
| CN109298646B (en) * | 2018-11-09 | 2020-05-05 | 珠海格力电器股份有限公司 | Scene control method and device, storage medium and electric appliance |
| CN109684916A (en) * | 2018-11-13 | 2019-04-26 | 恒睿(重庆)人工智能技术研究院有限公司 | Based on path locus data exception detection method, system, equipment and storage medium |
| CN109325474A (en) * | 2018-11-14 | 2019-02-12 | 郭道宁 | A kind of abnormal state detection method of couple of special caregiver of need |
| CN109584490A (en) * | 2018-12-10 | 2019-04-05 | Tcl通力电子（惠州）有限公司 | Safety protection method, intelligent sound box and security system |
| CN109558865A (en) * | 2019-01-22 | 2019-04-02 | 郭道宁 | A kind of abnormal state detection method to the special caregiver of need based on human body key point |
| CN109815921A (en) * | 2019-01-29 | 2019-05-28 | 北京融链科技有限公司 | The prediction technique and device of the class of activity in hydrogenation stations |
| CN110062210A (en) * | 2019-04-28 | 2019-07-26 | 广东安居宝数码科技股份有限公司 | Monitoring method, device, equipment and the storage medium of object |
| CN112183153A (en) * | 2019-07-01 | 2021-01-05 | ***通信集团浙江有限公司 | Object behavior detection method and device based on video analysis |
| CN110414360A (en) * | 2019-07-02 | 2019-11-05 | 桂林电子科技大学 | A kind of detection method and detection device of abnormal behaviour |
| CN110555394A (en) * | 2019-08-19 | 2019-12-10 | 西安理工大学 | Fall risk assessment method based on human body shape characteristics |
| CN110705505A (en) * | 2019-10-14 | 2020-01-17 | 国网山东省电力公司泰安供电公司 | Intelligent identification device, method and system for service specification of electric power business hall |
| CN111062356A (en) * | 2019-12-26 | 2020-04-24 | 沈阳理工大学 | Method for automatically identifying human body action abnormity from monitoring video |
| CN111062356B (en) * | 2019-12-26 | 2024-03-26 | 沈阳理工大学 | Method for automatically identifying abnormal human body actions from monitoring video |
| CN113705274B (en) * | 2020-05-20 | 2023-09-05 | 杭州海康威视数字技术股份有限公司 | Climbing behavior detection method and device, electronic equipment and storage medium |
| CN113705274A (en) * | 2020-05-20 | 2021-11-26 | 杭州海康威视数字技术股份有限公司 | Climbing behavior detection method and device, electronic equipment and storage medium |
| CN111652201A (en) * | 2020-08-10 | 2020-09-11 | 中国人民解放军国防科技大学 | Video data abnormity identification method and device based on depth video event completion |
| CN112017403A (en) * | 2020-09-15 | 2020-12-01 | 青岛联合创智科技有限公司 | Community-house integrated intelligent service electronic board |
| CN112382032A (en) * | 2020-11-13 | 2021-02-19 | 珠海大横琴科技发展有限公司 | Monitoring method and device, electronic equipment and storage medium |
| CN113177459A (en) * | 2021-04-25 | 2021-07-27 | 云赛智联股份有限公司 | Intelligent video analysis method and system for intelligent airport service |
| CN113505713A (en) * | 2021-07-16 | 2021-10-15 | 上海塞嘉电子科技有限公司 | Intelligent video analysis method and system based on airport security management platform |
| CN113705355A (en) * | 2021-07-30 | 2021-11-26 | 汕头大学 | Real-time detection method for abnormal behaviors |
| CN114419528A (en) * | 2022-04-01 | 2022-04-29 | 浙江口碑网络技术有限公司 | Anomaly identification method and device, computer equipment and computer readable storage medium |
| KR102591483B1 (en) * | 2022-11-16 | 2023-10-19 | 주식회사우경정보기술 | Apparatus and method for spatiotemporal neural network-based labeling for building anomaly data set |
| KR102613370B1 (en) * | 2022-11-16 | 2023-12-13 | 주식회사우경정보기술 | Labeling methods and apparatus including object detection, object tracking, behavior identification, and behavior detection |
| CN116430831A (en) * | 2023-04-26 | 2023-07-14 | 宁夏五谷丰生物科技发展有限公司 | Data abnormity monitoring method and system applied to edible oil production control system |
| CN116430831B (en) * | 2023-04-26 | 2023-10-31 | 宁夏五谷丰生物科技发展有限公司 | Data abnormity monitoring method and system applied to edible oil production control system |
| CN116434346B (en) * | 2023-06-12 | 2023-08-18 | 四川汉唐云分布式存储技术有限公司 | Method and device for detecting customer behaviors in unattended store and storage medium |
| CN116434346A (en) * | 2023-06-12 | 2023-07-14 | 四川汉唐云分布式存储技术有限公司 | Method and device for detecting customer behaviors in unattended store and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105718857B (en) | 2019-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105718857A (en) | Human body abnormal behavior detection method and system | |
| CN112257557B (en) | High-altitude parabolic detection and identification method and system based on machine vision | |
| CN106875424B (en) | A kind of urban environment driving vehicle Activity recognition method based on machine vision | |
| Alahi et al. | Socially-aware large-scale crowd forecasting | |
| EP2659456B1 (en) | Scene activity analysis using statistical and semantic feature learnt from object trajectory data | |
| CN101872418B (en) | Detection method based on group environment abnormal behavior | |
| KR101731461B1 (en) | Apparatus and method for behavior detection of object | |
| CN101800890B (en) | Multiple vehicle video tracking method in expressway monitoring scene | |
| CN107729799A (en) | Crowd's abnormal behaviour vision-based detection and analyzing and alarming system based on depth convolutional neural networks | |
| CN105787472B (en) | A kind of anomaly detection method based on the study of space-time laplacian eigenmaps | |
| CN108053427A (en) | A kind of modified multi-object tracking method, system and device based on KCF and Kalman | |
| CN104680557A (en) | Intelligent detection method for abnormal behavior in video sequence image | |
| CN108304798A (en) | The event video detecting method of order in the street based on deep learning and Movement consistency | |
| CN107330372A (en) | A kind of crowd density based on video and the analysis method of unusual checking system | |
| CN105303191A (en) | Method and apparatus for counting pedestrians in foresight monitoring scene | |
| CN105678803A (en) | Video monitoring target detection method based on W4 algorithm and frame difference | |
| CN103400391A (en) | Multiple-target tracking method and device based on improved random forest | |
| CN102930248A (en) | Crowd abnormal behavior detection method based on machine learning | |
| CN110532862B (en) | Feature fusion group identification method based on gating fusion unit | |
| Gong et al. | Local distinguishability aggrandizing network for human anomaly detection | |
| CN106778688A (en) | The detection method of crowd's throat floater event in a kind of crowd scene monitor video | |
| Zhang et al. | Safety helmet wearing detection based on image processing and deep learning | |
| CN104077571B (en) | A kind of crowd's anomaly detection method that model is serialized using single class | |
| Morozov et al. | Development of the logic programming approach to the intelligent monitoring of anomalous human behaviour | |
| CN106529405A (en) | Local anomaly behavior detection method based on video image block model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |